Today, Lattica is announcing support for NVIDIA cuPQC within the Lattica Platform and the HEAL runtime.
The integration brings NVIDIA GPU-accelerated cryptographic primitives into Lattica's encrypted AI execution stack, allowing developers to evaluate cuPQC on real Homomorphic Encryption workloads while continuing to use the same development workflow and APIs. A detailed technical walkthrough of the integration, including the software architecture, implementation, and performance evaluation, is available here:
GPU-Accelerated Cryptography for Homomorphic Encryption
Homomorphic Encryption enables computation directly on encrypted data, making it possible to build AI systems that process sensitive information without revealing it. While the underlying cryptography has matured significantly over the past decade, computational cost remains one of the primary barriers to practical deployment.
NVIDIA cuPQC represents an important step toward accelerating Homomorphic Encryption by providing GPU implementations of fundamental mathematical primitives such as the Number Theoretic Transform (NTT). By bringing hardware-accelerated cryptographic primitives to the NVIDIA ecosystem, cuPQC expands what's possible for privacy-preserving computing and encrypted AI.
Integrating cuPQC into the Lattica Platform
Within the Lattica Platform, encrypted AI workloads are executed through HEAL, Lattica's hardware abstraction layer for Homomorphic Encryption. HEAL lowers encrypted operations into cryptographic primitives and dispatches them to hardware-specific backends.
The cuPQC integration allows NVIDIA's accelerated computing to be introduced as another backend within this execution model, making them available to existing encrypted AI workloads without requiring changes to the compilation pipeline or application interface. One aspect we particularly appreciated during this work was the design of the cuPQC API, which fit naturally into HEAL's backend architecture.
Learn More
For a detailed look at the architecture, implementation, and performance evaluation of the integration, see the accompanying technical article:
Accelerating Encrypted AI with NVIDIA cuPQC
We are excited to continue working with the NVIDIA cuPQC team as the library evolves and additional cryptographic primitives become available.
